Use Deployment with leader election instead of StatefulSet

Deployment behaves better when a node gets disconnected from the rest of
the cluster - new provisioner leader is elected in ~15 seconds, while
it may take up to 5 minutes for StatefulSet to start a new replica.

deploy/kubernetes/statefulset.yaml → deploy/kubernetes/deployment.yaml

@@ -17,13 +17,15 @@ spec:
       port: 12345
 
 ---
-kind: StatefulSet
-apiVersion: apps/v1beta1
+kind: Deployment
+apiVersion: apps/v1
 metadata:
   name: csi-provisioner
 spec:
-  serviceName: "csi-provisioner"
-  replicas: 1
+  replicas: 3
+  selector:
+    matchLabels:
+      app: csi-provisioner
   template:
     metadata:
       labels:
@@ -34,8 +36,8 @@ spec:
         - name: csi-provisioner
           image: docker.io/k8scsi/csi-provisioner
           args:
-            - "--provisioner=csi-flex"
             - "--csi-address=$(ADDRESS)"
+            - "--enable-leader-election"
           env:
             - name: ADDRESS
               value: /var/lib/csi/sockets/pluginproxy/mock.socket

doc/development.md

@@ -0,0 +1,13 @@
+## Running on command line
+
+For debugging, it's possible to run the external-provisioner on command line:
+
+```sh
+csi-provisioner -kubeconfig ~/.kube/config -v 5 -csi-address /run/csi/socket
+```
+
+## Vendoring
+
+We use [dep](https://github.com/golang/dep) for management of `vendor/`.
+
+`vendor/k8s.io` is manually copied from `staging/` directory of work-in-progress API for CSI, namely <https://github.com/kubernetes/kubernetes/pull/54463>.